1. Field of the Invention
The present invention relates to a data search system and a data search method for flexibly and efficiently searching for intended data from a large amount of data.
2. Description of the Related Art
Conventionally, as a method for searching for intended data from a large amount of data, a method for conducting a search using text information describing data as a search key and a method for conducting a search using feature values in data as a search key are known.
The method using text information as a search key has the following problems. First, there is a problem regarding search accuracy. It is not easy to provide text information accurately representing data, so that text information provided to data is likely to be insufficient. In this case, it takes time and labor to search for intended data in search processing using text information specified by a user as a search key, which hinders enhancement of search accuracy. Second, it takes a long time for pre-processing. It is generally difficult to automatically provide each data with text information accurately representing the data; in most cases, it is required to manually confirm whether or not automatic provision of text information is appropriate, and it is also required to conduct an operation for manually adding text information, which results in a considerable time for pre-processing.
Therefore, the method for conducting a search using feature values in data as a search key has been paid attention to. In the conventional search processing using feature values as a search key, feature values used for a search are paid attention to, and feature values of each data are evaluated, whereby data having similar feature values is searched for.
However, in the conventional data search processing using feature values as a search key, there are the following problems.
It is difficult to convert human recognition into a numerical form together with a human personal point of view. Whatever feature values are used for specifying intended data, there are no feature values perfectly matched with a human sense. Therefore, it is difficult to determine optimum feature values for specifying intended data. Under such a circumstance, some feature values may be given as a search key by trial and error, whereby a data search is conducted for a trial. In this case, inappropriate data may also be included in search results. On the contrary, in the case where a search is conducted using text information as a search key, it is easy to find whether or not data contains elements having the text information as a keyword, and it is relatively easy to understand search results.
Furthermore, in the case where text information is used as a search key when a data search is continued to be conducted by trial and error, it is also relatively easy to assume text information given as a subsequent search key, based on the used text information and search results thereof However, according to the method using feature values as a search key, it is difficult to determine appropriate feature values for specifying intended data in a subsequent search, as described above. It is difficult to empirically determine which feature values should be given for conducting a search by trial and error.